Microsoft Frontier Tour London
Scale your intelligence on Windows with Surface - Kenneth Perry
Agents are growing at a increasing pace and when look at layers and foundation it starts with endpoint and with devices you put in people's hands to allow them to reach their full potential. Windows provides that control and security capabilities to provide that premium experience across the board and starts with trusted devices. Microsoft Agent 365 enables tools, and Windows is the execution layer for agents and Microsoft Agent 365 brings in things like Entra which identifies agents and Purview to mitigate threats and give control over capabilities to leverage and control agents. You can control things like with the OpenClaw Windows companion to control what it has access to and ensure that it has full control or does not have control, and these permissions can be controlled at every time.
Developers, developers, developers with Microsoft Foundry on Windows allow you to use local models, customise models or use different models. Surface is the endpoint solution for scaling hybrid AI where Surface is the physical extension of the Microsoft AI platform and enhances the experiences of people, to save time and adapt to how you want to work with security backed by Microsoft to build safe and secure when running agents on a device or be able to conduct business while out in the field without being connected to the internet.
You can unlock end-to-end transformation with Microsoft where can deliver unmetered intelligence to every desk and every home, think about the possibilities on how people can leverage technology to enhance their experiences, it is not a one size fits all situation where can match the right Surface to the work for worker knowledge with Surface Pro or Laptop, developer or advanced knowledge worker up to Surface Laptop Ultra or for an AI application developer up to Surface RTX Dev Box.
The endpoint matters more than ever, Surface and Windows are turning the latest devices into secure, workload-aligned AI infrastructure where can execute agents securely, build AI capacity, unlock AI with Microsoft stack and fit the device to the work.
Own your AI future with production-ready model customization - Jason Robey
Mentioned that their Uber driver got quiet when they mentioned that they work in AI, what they care about if got access to prescription drugs could they control them. Think about data and think about models you use and that you can't control so will talk about owning data. Fireworks are the leading open model inference engine and have zero-day access to latest models or half day if have to work through terms if that happens. Are doing 40 trillion tokens per day and 90% of tokens are on customised models for thousands of models. Their mission is to empower every company to own the intelligence that gives them their competitive edge.
Open-weight models are rivalling closed frontier models for a growing set of use cases, they are coming into parity with frontier models with thousands of workloads per work. Benefits of open weights drop price from 50% - 80% and performance tuned to your business and provide only capabilities you can ship. Fireworks on Foundry where are a first-party provider on Azure and get same cost as buying direct along with Azure security, terms and teams. Leading teams train models to make their own frontier to protect alpha, differentiate and lower costs, they have been able to move companies to open-source models with up to 90% saving on cost. They have Fireworks labs with PhDs who are ready to help you and get to a fine trained model on your data that you own.
Specialised intelligence maturity curve where start with open inference, the route to best models with nexus to get into IDEs and they have custom trained model to decide task complexity for developers to help them to know where to run things and know which models to switch to where can go from frontier model and have inference on open-source, then have people go to fine tuning where you turn your capabilities into a competitor advantage with specialised intelligence and can do full-weight training with your own evaluations and frontier training. They have had tasks that cost $8.50 down to $0.50 where your data is still yours where Fireworks is zero retention by default and you can download and run models somewhere else but can beat performance with customised solutions.
Keynote - Building UK's AI Frontier - Satya Nadella, Efe Abugo, Marlene Mhangami, Stuart Russell, Darren Hardman, Dame Caroline Clarke, Alex Kendall
Satya Nadella started by saying it was great to be in London and talk about broader ecosystem and how we move that frontier professionally and the firms we work with. Their mission is to empower everyone to achieve more and manifests in three pillars with the first where we need to recapture that sense of empowerment and take out mechanised parts of work, but how do you feel that sense of empowerment and how do you drive this and how do you put the human in the loop and ensure that sense of empowerment is not lost. The second challenge is where we have a digital system that is a learning system and you are continuously linking information to this system so as a business you need to capture and own the knowledge that is created including human and token capital. The third is trust, where have AI models that have more and more capability that is emergent, but as we speak scaling laws are creating more capable models that are less trustworthy so need t consider them as insiders and have zero trust with these black box systems as need to be accountable for these systems and that is what Microsoft are building out. Satya mentioned Copilot has all the models available to you for every form factor and every task, you can do the match between ergonomics of form factor, models and task available is the goal for building out Copilot and consider tokens coming from the PC where you can run fairly good models locally. They conceptualise this as a new type of operating system for work and that is what Copilot is all about.
Efe Abugo talked about an example where need to make a critical decision about expanding locally or globally and showed Microsoft Copilot app where can delegate complex tasks to Cowork were can use Code to do this and Autopilot to complete work and tasks on your behalf. With Cowork have choice or can choose Auto to have Microsoft choose the correct model. You can draft an email or perform other tasks and because Cowork is integrated you can open into Office such as a Word document including a confidential briefing and can get information filled in by others. Leadership is not going to make a decision on a document they will want to explore the document. Can describe the kind of application you want and Code will get you there including business context, when building with code it is a iterative process such as being able to add components to an application and then once created can see the application that code built which you can then publish and then share for the team to collaborate in. There is one more critical dependency where can bring in updated information, typically you would need to check for this but can delegate this to an autopilot that has its own memory, workspace and place in the hierarchy. This autopilot can be spoken to in a conversation and set up routines and then when information has been provided where the autopilot can follow up on your behalf and when shared can update things accordingly. You can go from an idea or plan or brief with Cowork, build an app with code and have autopilot work in the background enabling you and your organisation to achieve more.
Satya Nadella said that Copilot helping you with your tasks and in GitHub Copilot are making progress with features like HydraFusion with capable coding models where something can be done with a local model which provides an infinite token factory with auto models. There is also healthcare management on how you can improve patient outcomes such as in the NHS, you can build agents and have applied this in ambulance service to help get right service to the patients. Microsoft shared a video where showing the NHS has an opportunity to deploy the best options to help provide care and sustain NHS, where have calls that can generate many data points and if can use this better can provide better value, data can be across multiple sources and Copilot can bring together multiple data sets and other ways to help and provide the best care and patients can be vulnerable so advocating for best care is important.
Satya Nadella stated that Microsoft IQ has a couple of different parts where have action space registered such as connectors, skills and plugins for agentic system are available in a safe registry and then the next is a rich context where can structure what's underneath Microsoft 365 which is rich from communications data, such as email or Teams or dashboards you may have created in PowerBI or knowledge bases with Foundry IQ as well as memory being built but you also want to use multi model in your agentic system and have keys to separate harness from models to keep actions space, context and memory outside of the model space so if you can't remove a model then could you still continue, if you can't then you're bound to that model, what would happen if it went away so this is consideration when deploying sophisticated AI models. Microsoft Decision-1 enables which model to choose for a task which allows you to pick right model and make this more efficient.
Marlene Mhangami talked about someone who has just started so build something to help with business processes created in GitHub Copilot and using Microsoft Foundry which can use the skill for this to allow the context to be used to build a Foundry AutoPilot called Mona to support their team, where start with a prompt and then to create an agent and then can build agent where decision models are built for frequent classification such as Microsoft Decision-1 and can also include other models and features with toolboxes. It is then possible to try out the agent locally, they usually check in with their team every two weeks to want their agent to do the same things in the same way and an agent can use a tool for memory including scopes and recall to get information and then pull in updates from Teams and Outlook and have the decision model whether things are being completed and what should be checked in on, you can once the agent is working you can check the code is following best practices and validate this and get a validation report where can get things including what to improve and then can make those changes and then deploy this to Foundry for use in Teams, Outlook or anything else you can deploy the endpoint into. Agent can use decision model to see if it can be helpful in specific case and can work with surfaces it is using and you can evaluate how it is performing from traces which can be complicated so there is an insights option to give ways to optimise agents so they improve over time.
Satya Nadella talked about partners building out specific use cases with agent systems from flights to insurance which includes building an agentic system in the first instance, or others to bring in orders into a system or replace a manual process to increase throughput and reduce burden and it is great to see ambition level and real work impact of agentic systems. Building trust is important with Agent 365 and building principals of observability and governance where need to id agents and log their activities including tool cools, chain of thought and logs have to be tamperproof and be able to set policy to contain or monitor actions. Observably at every layer the stack including token budgets and when you collect observability data you can use this to go back into optimisation and improving agentic system to continuously improve the system. Perception is you need to manage systems with an agentic system of its own that is driving security such as red agents to simulate attacks, blue agents which detect and green that fix and remediate issues. Project Perception allows dynamic agent risk, dynamical conditional access and dynamic permissions with guardian agents and playbooks.
Satya Nadella also mentions token cost per watt is important. Microsoft have over 500 data centres and have two live and a third coming online soon in the UK. Upskilling is important where have upskilled over 2.5 million people in the UK and other element of infrastructure is silicon diversity at scale from Nvidia, AMD and their own such as Maia 300 and look forward to bringing innovation to deliver the performance frontier. Sovereignty is important to ensure you have the control, are trying to manage risk and best way is to have portfolio approach so a combination will give choice of controls and resilience and just announced many features this week including ability to have clean rooms to have ability to detect any tampering of logs during administrative process and give full disconnected operations in your private cloud and have announced a new control plane to manage this with Sovereign Public and Private Cloud.
Stuart Russell talked about sovereignty which means different things, the Sovereign Control Plane is a unified view which enables you to see what is going on such as data residency, data confidentiality, business continuity and can see the percentage of where resources are so can investigate out of region resources and have these redeployed where needed. Can you keep operating if access to public cloud is disrupted so need extra confidence to operate in extreme circumstances so with Sovereign Business Continuity to allow from Azure public to Azure local and show resources that are essential to business that would require changes to move a specific resource from Azure to Azure local what is needed such as different messaging service to allow business to operate on Azure local when necessary. You can manage resources on Azure Private or Local in your Azure estate and can disconnect from Azure and can disconnect completely without disrupting your estate. You can bring models and developer tools along with workloads and data as sovereignty is not a one size fits all so can have choice and control to meet these requirements.
Satya Nadella concluded by saying there were a lot of systems that need to work together to enable Sovereignty on Azure and that ability is fantastic to see. You don't want to sacrifice ability to protect this when move to local data centre where cyber resilience is important and need to ensure this global protection and intelligence with local control and ensure cyber security even for things that are disconnected or local. This is the task they are building to fundamentally build systems to help people and enhance learning to deliver sophisticated systems.
Darren Hardman talked about how the UK will complete and bending the curve will be on how much UK can achieve and see that progress in people's everyday lives and the UK is eighth in world for generative AI adoption with Ireland in third, but lesion is deliberate action. They spoke to the Prime Minister about security and reliance that must underpin it all and we have a role to play in AI safety and will be taking on the G20 presidency and can say how most powerful AI should be governed. Only two in five employees are confident about AI and challenge is to optimise strengths in each company, and regions to empower people and organisations to achieve more. Adoption isn't enough and have to build trust, question is not if you have a strategy about AI but if it gets work done and for the UK AI opportunity reaches into the industry of everyday life. AI can help Network Rail make better use of their vehicles and agents can put better and more timely information into businesses. With the AA thanks to AI, 18,000 technical documents are quickly and readily available to AA or have thousands of Openreach engineers use copilot of with BT to enable teams to use Copilot to support their work, it can help make them more confident in their work and do things quicker and open up opportunities they felt were off limits to open new possibilities for people. Microsoft Azure and AI can help find answers and documents or emails and help training for UK Power Networks where AI becomes part of how business works. People need skills to use AI infrastructure and Microsoft committed $30 billion to the UK and now have a next gen facility for AI and cloud in Acton and part of wider network and there is a new datacentre in Newport in Wales and other sites across the UK. There are few places where things are more critical is healthcare to help staff spend less time on admin to help bring more care to patients and a trial with Copilot showed a saving of 43 minutes per day and give more time to patients and support for people.
Dame Caroline Clarke is CEO of the Online NHS Trust joined Darren and spoke about their aim is to provide more options digitally and spoke about what AI can do for patients. People like them are using Copilot and people are using AI for care coordination including simulating care and predicting care, using AI in diagnostics which is not universal but will be more in the future. Having an AI strategy is one part but making it happen is the most important so how do they manage change, all digital projects are change projects and have Copilot Cafes to share and have run continuous improvement programmes and if people get things wrong that is okay as can learn and it needs to be clinically led and having flatter structures helps and the Copilot Café has over 100,000 users. How to ensure appropriate guardrails and in healthcare it is all about trust and is highly regulated and at a point where need to gain trust not assume trust and people are turning to large language models for advice that may not be correct, so do you trust that technology to look after yourself as a person.
Darren Hardman talked about how NHS is embracing AI and AI is building opportunities to use AI to generate growth and serve customers and helping them succeed. Work IQ brings context across emails and documents which can help products such as Wordsmith from Edinburgh which enables in house legal teams to use AI and serve legal teams at scale. People are using AI across the UK from places like Newcastle with Origin to use AI to solve real problems here in the UK along with Wayve bringing intelligent self-driving with potential to improve millions of lives.
Alex Kendall from Wayve talked about previous versions of robotics which had hand crafted rules but with self-driving need to deal with situation it has never encountered before and are launching a service in London with Uber with Uber X can have a chance to have a robo taxi ride with a supervisor driver. Main difference is there are clear decisions to get right and need to work on mass market hardware with an efficient system that can assess risk and need to build a simulator and reward model to ensure best possible outcome. Main challenge is do you build your own AI system or get one off the shelf but the ecosystem they are building is what is needed to leverage frontier AI with an ecosystem play and what they can access and customers building thousands of cars so helps get more information and data which is going to define the frontier and can;t have same AI but can provide frontier capabilities and then work with brands to innovate such as choosing compute, shape and design or car, placement of sensors, the driving experience and behaviour and tune these systems for the experience those brands want. Microsoft partnership has helped them move from a startup which was compute hungry which allowed them to train models on Azure with hundreds of gigabytes of data, and the ecosystem is predicated on strong infrastructure with trust along with innovation.
Darren Hardman mentioned the question is how to we change trajectory of what the UK can achieve, for every organisation the challenge is to turn AI potential into improvements people can feel including services and opportunities and Microsoft brings intelligence and trust together. Bring human approval where it matters but when agents take on more responsibility can take action or saw what they did and remain in control. Giving more people skills and opportunity to take part with more AI skilling and want AI to open doors whatever their background or where they live and make sure opportunity reaches this and make UK where breakthrough in research helps move things forward and deliver confidence to help more people participate, bring shared ambition to turn opportunity to progress.
Build agentic AI Postgres apps on Azure HorizonDB - Josh Ndemenge
Josh talked about the session which would cover how AI can make better decisions with a compliance advisor to understand business context, retrieve trusted information, connection relationships across data sources and ground responses in evidence and recommend actions while keeping responses readable. Will have a user who needs to check for compliance issues, see if there is a precedent, is there a risk and can the business proceed. The knowledge lives in Azure Cosmos DB which is conversations, user context, recommendations and will provide the application with memory, also use Azure SQL Database Hyperscale for regulatory reviews, audit history and compliance workflows.
Azure Horizon DB is for legal cases, policies, graph relationships and semantic search. What do they need to know, are asking “have we encountered a similar compliance issue before?”, “What evidence supports that conclusion” and “What are the consequences if we get it wrong”. Will be building a compliance advisor where can ask a single question and get an answer. Azure Horizon DB is just PostgreSQL, but it has more functions built in with AI functions and pipelines and is up to 3x faster transactional processing and vector search than self-managed and is zone resilient with auto-scaling storage up to 128TB.
Azure Horizon DB is 100% PostgreSQL compatible and can for any read/write it can go through primary but have separate load balanced endpoints across replicas so don't have to have all reads go through one point and storage uses Azure Blob Storage where main advantage where storage can grow independently from compute so can distribute workloads as usage grows. Can build an agentic app that uses Azure OpenAI and Mem0 to capture and retrieve memories.
Delivering Sovereignty in the Age of AI - Alistair Speirs & James Ancell
Alistair talked about the thing they have heard the most is what the heck is going on, it is a crazy time, there has been radical change before with the internet or the fall of the Berlin Wall, is AI conscious, is it human or a card trick using statistics some of you probably have CDs with older software. There is also an inflection point in geopolitics with benefits but also those who have been left behind and the fairness of the system, there is another timestamp like Brexit which made other challenges that put the EU in different path to the UK and having to deal more and more with these geopolitical tensons.
Sovereignty is on every agenda but rarely defined. In IT we are designing for multiple outcomes at once with Security to guard against uncontrolled or unauthorised access to your data and infrastructure, resilient where can keep essential services running through shocks from cyber attach to outage to network disconnection and sovereign which is exercising control over your data, assets and choices including in event of foreign intervention. Location is not the same as control as governments put transparency, compliance and access controls well ahead of where infrastructure is. Sovereignty is the capacity to choose and control where data is stored and processed, which operation and governance procedures apply to data and AI and the UK tends to take a cloud forward approach.
Digital sovereignty requires balancing risks with business continuity for how to maintain operations, cyber resilience and security to protect against evolving cyber threats, digital sovereignty requirements with how to interpret risk and make defensible decision and strategic innovation requirements but biggest risk to companies in business is going out of business. You can't survive without a strong approach to digital sovereignty and there is a risk framework for this to assess the risks together which could range from different things from climate change to energy independence.
Microsoft Sovereign Cloud enables not just a binary choice as risks change over time where low probability risks can move to a high probability risk and something may have a low impact and as it develops and becomes more mission critical it becomes higher so need to build a way to be flexible on how you respond. There is a Sovereign Public Cloud but you may want to take Microsoft out of the equation and go to Sovereign Private Cloud and if design applications using Azure Arc have high portability into regions around the world or privately and support multi tenancy infrastructure including infrastructure from another provider for various reasons from low latency to a capital expenditure requirement where developers build for this platform but don't have to make that decision on where this should be which should be a business decision.
Microsoft has contingencies and expertise to manage legal requests and are committed to complying with all applicable laws and do not comply with requests that are not unlawful and will challenge these where needed to defend your data and build contingencies into products where can have advanced forms of encryption to control keys and can quickly move data around world if situation changes or need to. Cloud resilience and business continuity allows for connected and disconnected scenarios plus can replicate and fail over to a recovery region you choose and support multi-country failover and resilience.
Powering AI with Sovereign Private Cloud is challenging where to have minimum workload could be using a different model so customer model choice is important and can benchmark these in advance. Foundry Local Model offerings support Open Source Models along with Frontier Models for qualified customers and if have to ask how much that costs you probably can't afford it. Sovereign landing zone allows you to optimise the approach which can be complex but there are scripts that make it simpler. UK policy already leads the way where data can be outside of the country if needed and the policy of the government is for public cloud first, where control comes from requirements not labels for technology, data, operations, supply chain along with commercial and governance.
James Ancell joined to talk about things including how AI is discussed in Government where they have different levels of interest and excitement so Department of Innovation, Science and Trade had a lot of knowledge but had risk of shadow AI as they want to use the latest technology but now in the Department for Business and Trade there is still a high appetite but the difference is more public facing services and trade is more of a cyber target means a different risk appetite but a lot of optimism. There are jurisdictions need to work with, so how does sovereignty play in this? Where dealing with different layers and think about things in terms of scenarios and how they may play out in a less linear way. Issue with risk registers sometimes gives a false impression of the risk, things can play out in combinations and if look at different risks may have different conversations where can delegate risk but not the scenarios, you can stress test your plans and Microsoft's plans to enable you to stick to your plans, don't wait for the disaster to happen but plan it out with Chaos engineering and can build from this, you can identify the signals you would be seeing that would impact things or things to be looking for, so these can be things you can monitor to see where things are going wrong.
James Ancell mentioned everyone is building markdown files, then single page HTML files and then need somewhere to have these files and internally in Microsoft had different systems so what was favourite cultural thing with AI that has emerged is, there is a debate going on with vibe coding but people can make agents that are poorly labelled and have usability issues, citizen developers will build things without rethinking so do we try and control this by monitoring and police this or is this an education piece about how you can build things safely. People can be proud of something that does something cool that may be exciting to them but terrify those in the industry. What would expect to see change with AI in the future, but the bit that excites them with government and AI where can make a lot of this invisible such as renewing passports and make this really easy or don't need to monitor latest changes in legislation but integrate this into your training tools, AI will solve some of the supply issue but will be a real dilemma. Microsoft has experts all around the world who can help. Many things take a decade to be noticed so the challenge is to think about risks and opportunities now to bring foresight for the next ten years.
Empowering AI Adoption through Skills Development - Stephanie Neale
Stephanie Neale talked about AI transformation means workforce transformation and will show that skills take AI ambition into AI adoption where can help employees succeed. Where need to respect the learner and respect the opportunity where fear is what is holding people back but in that fear zone it can be hard to learn but taking that person on that AIK skills knowledge means courage turns to confidence and is about that mind shift change, not just about the AI tools. You need to think about where you are as an organisation and don't over complicate things, ask AI how it can teach you to make it do better and that learning can become real and leaders can help those to take that first easy step.
Workforce success matters where people need a strong foundation, needs to be role specific, with human skills, part of leadership development and have real world application and capabilities will apply where one group will matter the most. Human skills is curiosity which has transformed humanity from the very beginning, then courage which helps move forward to re-imagine work and not fear losing it, compassion where can lead with empath and put people first, creativity where can move boldly and learn as you go where ideas move to something tangible and collaboration to create right environment for colleagues to learn from each other. It takes human skills to get to performance needed and learning needs to be continuous and organisations that embrace the cultural shift of this can outperform competitors by up to 68%.
Microsoft's workplace transformation has three connected areas with workforce development where fear is acknowledged openly with dedicated time to learn and learning in performance goals, then there is talent retention with a career hub allowing people to expand beyond their role by taking projects outside of that role, next logical skills and intentional career planning and employee experience with AI tools that reduce toil, rigorous onboarding plus rewards and recognition, with a high commitment to supporting career development and make sure technology works for people in the business to reduce toil and take away repetitive tasks while making AI onboarding interesting and deliver that productivity.
Microsoft applied practices at Microsoft to drive impact including Sales with training tailored to roles, gamified experiences such as agentathons with agentic AI coaching. Customer Service with community of experts with learning champs along with peer-to-peer learning which they treated like a project, team leader engagement to track measure and adjust to figure out where this wasn't working. Engineering helped with dedicated learning time, show-and-tell sessions with collaboration together where technology was evolving so quickly and skills verification. Giving people a save place to practise and a community where skills and knowledge can grow quickly. Structured skilling accelerates measurable value where can deliver improved employee productivity, faster business value realisation and reduced internal skilling effort as not as much may be needed in house. Start with what AI means to employee and the rest can follow from there and let people learn or grow and give them that time back.
Microsoft skilling differentiation is how they work internally and how they support their customers where have AI powered learning experiences which is AI-customised and verified, role-based skilling and organisational tooling plus is human focuses where have employee skilling, custom skilling plan for human skill building and delivering, AI enables scaling using AI but human delivery where context, connection and trust matters to support acceleration and adoption of AI.